#neural-odes

Mixed Precision Training Stabilizes Neural ODEs

Researchers demonstrate a framework that reduces memory use by 50% and speeds up neural ODE training 2x by carefully mixing low and high precision arithmetic.